HypoArena: leaderboard
HypoArena (2026) evaluates prospective hypothesis discovery: from inconclusive evidence reconstructed out of expert documents, a model proposes grounded, testable hypothesis sets that a seed-2.0-pro judge compares pairwise, with the source-derived reference set as an anonymous competitor. The board is the average of six domain Bradley-Terry-Davidson ratings, anchored at 1500 with no fixed maximum; higher is better.
Metric: Average of the six domain ratings: Bradley-Terry-Davidson rating (base 1500, open-ended) from reference-independent pairwise judgments by seed-2.0-pro, both presentation orders, with the source-derived reference hypothesis set as an anonymous competitor; baseline single-pass generation; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 (High) | 1654 | #161 (Claude Sonnet 4.6) |
| 2 | Claude Opus 4.6 (High) | 1652.7 | #102 (Claude Opus 4.6) |
| 3 | GPT-5.4 (High) | 1631.4 | #123 (GPT-5.4) |
| 4 | Kimi K2.6 | 1582.7 | #178 |
| 5 | GLM-5.1 | 1581.9 | #179 |
| 6 | DeepSeek V4 Pro (High) | 1535.4 | #188 (DeepSeek V4 Pro) |
| 7 | MiniMax-M2.7 | 1498.9 | #518 |
| 8 | DeepSeek V4 Flash (High) | 1484.6 | #258 (DeepSeek V4 Flash) |
| 9 | GLM-5 | 1464.3 | #260 |
| 10 | MiniMax-M2.5 | 1423.8 | #582 |
| 11 | GPT-5.4 Mini (High) | 1421.9 | #389 (GPT-5.4 Mini) |
| 12 | Gemini 3.1 Pro (Preview) (High) | 1394 | #94 (Gemini 3.1 Pro (Preview)) |
| 13 | Gemini 3 Flash (High) | 1345.7 | #164 (Gemini 3 Flash) |
| 14 | Kimi K2.5 (Thinking) | 1289.5 | #256 (Kimi K2.5) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hypoarena · How It Works · Data refreshed daily, snapshot 2026-11-13.